WORKING PAPER 66December 2018

Priming the Pump: Does Aid Pave the Way for Investment?

Samuel BrazysSchool of Politics and International RelationsUniversity College Dublin

AIDDATAA Research Lab at William & Mary

Abstract

Recent advances in the coverage and precision of development data have opened exciting new avenues for analyzing the political economy behind the allocation and effectiveness of foreign aid. While a number of recent papers have looked at the social, environmental or welfare implications of aid, this paper instead focuses on how aid impacts an intermediate outcome in the development process, foreign direct investment (FDI). Combining geo-referenced data on foreign aid with a similarly coded dataset of FDI project locations, the paper uses a quasi-experimental, spatial-temporal, identification strategy to evaluate if the location of foreign aid projects presages later FDI. Drawing on the literature on the political economy of aid, the paper develops theoretical expectations about a given donor’s aid and FDI projects from that state before finding that local aid increases the chance of a location attracting FDI by up to 40 percent. While aid and FDI from bilateral donors goes hand-in-hand, both Chinese and EU aid also attracts FDI from other sources.

Author Information

Samuel Brazys School of Politics and International Relations University College Dublin Dublin, Ireland samuel.brazys@ucd.ie

The views expressed in AidData Working Papers are those of the authors and should not be attributed to AidData or funders of AidData’s work, nor do they necessarily reflect the views of any of the many institutions or individuals acknowledged here.

Acknowledgements The author thanks the European Union’s Horizon 2020 research and innovation programme under grant agreement number 693609 (GLOBUS) for generous funding support for this project. Comments/critiques/suggestions welcomed.

1

Introduction

The geo-spatial revolution in development data has facilitated the rise of new research

agendas which consider the political economy of subnational and local aid allocation and

impact. Recent studies have used geo-referenced data to consider questions of aid’s local

impact on topics including growth, welfare, the environment, and governance (Dreher and

Lohmann 2015, Bizter and Goren 2018, Blair and Roessler 2018, Martorano et al. 2018).

Other work has considered the political motivations behind sub-national allocation of aid

(Briggs 2017). This paper also takes advantage of precisely-located aid projects to

investigate if and how aid can serve as a precursor for foreign direct investment (FDI). While

a number of papers have evaluated the aid-FDI relationship at the cross-national level, there

are few that theorize or test the relationship at a local level. This paper contends that aid can

often lay the local groundwork for FDI by providing the physical, if not institutional and

human, infrastructure that is needed for private enterprise. However, these expectations are

tempered by theorizing that these efforts may not be universal in their effects, but instead

might be structured so as to facilitate source FDI primarily from the donor country. Donors

that are more explicit about the strategic economic or political aims of their aid programs

may be more likely to have projects that attract their own FDI.

In order to evaluate these claims, the paper utilizes several geo-coded foreign aid datasets

in Africa from the AidData project combined with nearly 10,000 geo-referenced FDI project

locations from the Financial Times fDi Markets database. These geo-referenced data allow

employment of a spatial-temporal identification strategy that uses information on project

timing and location to construct a quasi-randomized environment in which one can observe

an aid treatment effect. The analysis follows a difference-in-difference approach that

compares locations with active aid projects at the time of the first FDI project those those

that do not have an active aid project, but subsequently will. Both of these sites are then

compared to locations that have no aid project throughout the duration of the dataset.

The results suggest that, in general, aid is substantially effective in attracting FDI, with

locations with active aid projects up to 40 percent more likely to receive a subsequent FDI

project compared to sites where aid projects are not yet active. However, when looking at

aid and FDI from individual source actors, the results are more nuanced. Local aid projects

from a given donor are extremely likely to be co-located with FDI from that actor, but the

difference between an active and inactive site is negligible, suggesting simultaneity rather

than causality. Yet, active aid from some individual donors, notably China and the EU, does

2

also appear to attract FDI from other countries. In contrast, aid from Japan and the US has

no impact on attracting outside FDI.

Aid and FDI

A substantial amount of literature has suggested that aid may precede or facilitate FDI,

including in Africa (Anyanwu 2012, Amusa et al. 2016), although aid may also

simultaneously serve to crowd out FDI (Selaya and Sunesen 2012). Foreign aid can boost

economic infrastructure (Donaubauer et al. 2016), serve in a signaling function, especially in

post-conflict countries (Garriga and Phillips 2014), or facilitate human capital and social

cohesion (Donaubauer et al. 2014, Cleeve et al. 2015, Addison and Baliamoune-Lutz 2016)

which in turn attracts FDI. While most studies have suggested that the relationship between

aid and FDI is positive, some have suggested this only applies for some countries (Kimura

and Todo 2010, Arazmuradov 2015) while other work has found a negative relationship

between the two (Donaubauer 2014). However, with the notable exception of Blaise’s (2005)

study on Japanese inflows to China, nearly all of this literature focuses on the relationship

between ODA and FDI at the recipient country level.

As shown above, the cross-border literature suggests several pathways by which foreign aid

can induce investment. When considering the link at the local level these causal

mechanisms become clearer. A variety of different aid projects may increase local suitability

for FDI. Most obviously, aid classified as “Aid for Trade” (AfT), may help improve local

business conditions and attract FDI (Lee and Ries 2016). AfT is a broad classification and

includes categories of productive infrastructure – including transportation, energy,

communications and utilities infrastructure – but also can include, often industry-specific,

technical training or research and development (Brazys and Lightfoot 2016). The locational

link is most obvious with physical infrastructure, but it is also plausible that aid projects which

upskill local labor pools will also make that location more attractive to FDI (Donabauer et al.

2014).

The existing literature is relatively agnostic regarding heterogeneity in the aid-FDI

relationship, usually neglecting to consider if the source of aid and/or FDI may contribute to

its linkage.1 However, there is substantial reason to suspect heterogeneity in the aid-FDI

relationship depending on the political economy of the source country. Considerations of

foreign economic policy motivation date to at least McKinlay and Little (1977) and have

1 Kimura and Todo (2010) who find that Japanese aid only attacts Japanese FDI being an important exception.

3

sustained a prolonged debate if foreign aid is given to suit “donors’ interests” or “recipients’

needs” (Alesina and Dollar 2000, Berthelemy and Tichit 2004), or, more subtly, if it is

“targeted” for development purposes in countries most likely to engender spillovers to the

donor (Bermeo 2017). Extending the logic of this literature, it is eminently plausible to think

that aid from some donors may be used to facilitate FDI from that country.

While there is reason to suspect general country-level heterogeneity in the aid-FDI

relationship, the empirical literature also gives clues as to country-specific expectations. As

the largest foreign aid donor, the United States has long been pilloried as self-interested

(McKinley and Little 1979), although the empirical findings are mixed as to the extent to US

engages in purely egotistical aid allocation behavior (Brazys 2010, Harrigan and Wang 2011,

Bearce et al. 2013). Likewise, the second-largest historical donor, Japan, has historically

faced accusations that its aid program is driven for geo-economic reasons (Hook and Zhang

1998), particularly since official documentation is often naked in these aims and indeed

Japan has been criticized for use of tied aid as an extension of “Japan Inc.” (Hall 2011).

Indeed, both country-level and sub-national studies have suggested that (only) Japanese

FDI follows Japanese ODA (Blaise 2005, Kimura and Todo 2010). In contrast, the EU is

often portrayed as a “normative power” driven by more altruistic aims, although, again, the

evidence here is mixed (Carbone 2013, Brazys 2013). Indeed, in the case of the EU, this

may be driven by the internal incoherence of the EU’s “shared competence” with member

states in development policy.

Outside the traditional donors of the OECD’s Development Assistance Committee (DAC),

China has recently emerged as a major actor in the development space. Once again, there

is a popular perception that Chinese development efforts are primarily intended to help

China, and indeed there is some evidence that Chinese aid flows increase Chinese FDI

flows to the same country (Su et al. 2017). However, the empirical evidence is again mixed,

with some work suggesting that Chinese aid is effective in boosting growth (Dreher et al.

2017), while other work suggests Chinese development efforts may undermine local

governance (Brazys et al. 2017, Isaksson and Kotsadam 2018) or traditional donors’

conditionality efforts (Hernandez 2017). To shed light on this heterogeneity, the empirical

section below not only examines the general relationship between aid and FDI, but also

considers actor-specific relationships for the US, EU, Japan and China.

Data and Methods

4

The primary outcome variable is greenfield and expansion FDI projects drawn from the

Financial Times fDI markets database, which has been used in a number of recent studies in

economic and political science (Gil-Pareja et al. 2013, Brazys and Regan 2017, Owen

2018). This data includes not only data on project characteristics (size, sector) but also

geographic data on source country and project location. This study considers 9,864 projects

in fifty-six African countries from 2003 to 2017. Of these, 6,133 project records contain

geographic destination at the city-level, and accordingly the analysis uses these to identify

local effects. While the data does include information on project size, both in terms of

investment amount and job creation, the bulk of this is estimated, and potentially biased.2

Accordingly, this paper relies on project counts as this data is verified and cross-referenced

in the original fDi Markets methodology.

The primary explanatory variables come from AidData’s geo-coded datasets. In the primary

analysis, the paper relies on seven country-level “Aid Information Management System”

(AIMS) datasets from Burundi, Democratic Republic of the Congo, Malawi, Nigeria, Senegal,

Sierra Leone and Uganda (Peratsakis et al. 2012; AidData 2016a; AidData 2016b; AidData

2016c; AidData 2016d; AidData 2017a; AidData 2017b). These datasets capture geo-

referenced aid projects from most OECD donors as well as the World Bank. This data is

combined with similar project-level, geo-coded, data on Chinese development efforts

(Strange et al. 2017). The primary analysis considers projects in that data coded as “ODA-

like.” In the robustness checks, Africa-wide data for both Chinese and World Bank aid is also

used. Collectively, these data cover 2,633 projects at 4,315 locations from 2000 to 2014.

Map 1 displays the spatial locations of the AfroBarometer respondents (purple stars), ODA

projects (white squares) and FDI projects (gold circles).

In order to conduct the empirical analysis, the paper takes advantage of both spatial and

temporal dimensions of the data in order to employ a quasi-experimental, difference-in-

difference, approach similar to that used in Knutsen et al. (2017). First, the paper uses

spatial information to identify sites that are in proximity to aid and/or FDI projects. In the

primary analysis, the paper utilizes enumeration areas from the geo-coded AfroBarometer

surveys as the site locations (BenYishay et al. 2017). AfroBarometer sites have the distinct

advantage of facilitating the use of site-specific governance variables, in particular local

experience with and perceptions of corruption, both of which have been shown to be

correlated with local FDI (Brazys and Kotsadam 2018). The caveat of this approach,

2Brazys and Kotsadam (2018) who also use this data find a significant difference in the total amount of FDI calculated via

the fDi Markets data when compared to official World Bank statistics.

5

however, is that the location of Afrobarometer enumeration sites could be endogenous to aid

or FDI project siting. As such, the interpretation of the main results is a comparison of the

likelihood that aid attracts FDI at Afrobarometer sites. To address this caveat, and look for

more generalizable impact, grid-cells are used as the base locations in the robustness

checks below.

Map 1: FDI, Aid and AfroBarometer Respondent Locations

The analysis next takes advantage of the fact that both the aid and FDI project records

indicate the timing of projects. This information is used to identify site locations where an aid

project was active at the time the first FDI project began as well as inactive sites where an

aid project would begin after the first FDI project. Both of these sites are then compared to

sites with no aid project at any time. This approach enables a mitigation of endogenous

selection effects, that is that there is some additional characteristic(s) of a particular site that

makes it attractive to both aid and FDI. By taking a difference-in-difference between active

6

and inactive sites, the paper can evaluate the impact of an aid project “treatment” on the

likelihood that a site attracts FDI. In reduced form:

(1) 𝑌𝑖𝑡 = 𝛽1 ⋅ 𝑎𝑐𝑡𝑖𝑣𝑒𝑖𝑡 + 𝛽2 ⋅ 𝑖𝑛𝑎𝑐𝑡𝑖𝑣𝑒𝑖𝑡 + 𝛼𝑠 + 𝛿𝑡 + 𝛾 ⋅ 𝑿𝑖𝑡 + 휀𝑖𝑣𝑡

where the first FDI outcome Y for site location i at year t is regressed on dummy variables

active and inactive for aid projects at the time of the FDI outcome. The models below control

for both country and year fixed effects and use robust standard errors. The models

below also control for a vector ) of site-level control variables aggregated from individual

responses the Afrobarometer. The baseline set are perceptions of corruption of government

officials as a measure of local governance quality and an urban/rural indicator for the

respondent site. The results are checked for robustness against no controls and expanded

controls below.

Analyses using a spatial identification approach similar to the one here have to make an

assumption about the geographic reach of the treatment. This is ultimately an empirical

question that includes a trade-off between the precision of the geo-location in the data,

noise, and the size of the treated unit, which in this case is a polygon. The analysis employs

precision code “2” in the AidData, which is “city-level” like the FDI data, or precise to roughly

25km. Accordingly, a cut-off off less than 25km is not justifiable given the precision of the

data. Most studies using this approach have settled on 50km as the ideal distance for

evaluation (Knutsen et al. 2017, Brazys and Dukalskis 2017, Brazys and Kotsadam 2018)

and the primary analysis below uses that distance. However, 25km and 75km treated

distances are also examined, with the expectation that there is a dilution of the treatment

effect as the distance is increased.

Results

The results from the combined model are available in Table 1. The hypothesis that aid

attracts FDI is strongly supported. In the 50km Model (2), sites with an active, proximate, aid

project within 50km are 27.1 percent more likely to attract at least one FDI project compared

to sites where there is no active aid project but that will eventually get an aid project. This

magnitude is quite substantive and is strongly suggestive that aid presages FDI. The

difference-in-difference is significant at the 1% level. The negative sign and significance of

the coefficients, particularly on inactive, is also of interest, suggesting that sites which will

receive aid projects are less likely than non-aid sites to receive an FDI project. One

7

interpretation of this result is that aid goes to otherwise “disadvantaged” sites, suggesting

that it is appropriately targeted to areas of need. As the coefficient on active is still negative,

it suggests that aid does not completely eliminate this disadvantage, but does reduce it

significantly. The 25km (Model 1) and 75km (Model 3) results are consistent with the

expectation that the treatment effect reduces in distance. While difference-in-difference for

the 25km result is both of a larger magnitude and increased statistical significance, the

difference-in-difference for the 75km result is smaller and not statistically significant.

Table 1: All Donors in African AIMS Countries

(1) (2) (3) VARIABLES 25 km 50 km 75 km

Active -0.038 -0.137* -0.377*** (0.052) (0.072) (0.066) Inactive -0.447*** -0.408*** -0.318*** (0.125) (0.073) (0.075) Observations 1,371 1,371 1,371 R-squared 0.262 0.234 0.259 Baseline controls YES YES YES Year FE YES YES YES Country FE YES YES YES Difference in difference 0.409 0.271 -0.059 F test: active-inactive=0 9.244 7.191 0.364 p value 0.002 0.007 0.546

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

While the general hypothesis is unambiguously supported, the source-actor results reveal

more nuance. When considering the impact of source actor aid on source actor FDI (Models

1, 3, 5 and 7) there is no increase in FDI for active aid sites compared to inactive sites.

However, in all cases, there is strong correlation between aid and FDI projects from a given

source actor at a given site, as seen by the large and positive coefficients on both active and

inactive. In all instances, the probability that a site with active or inactive aid from a source

actor also gets an FDI project from that source actor is at least 0.56, and in the case of

Japan (Model 5) and China (Model 7), around 0.8. This is strongly suggestive that aid and

FDI from a given source go hand-in-hand, with neither causing the other. Instead, a plausible

interpretation is that donors focus on particular sites and direct both aid and investment to

that area. There is little variation across the donors in the outcome, although the correlations

are slightly larger for both Japan and China.

Interestingly, however, in two cases aid from a given donor does appear to increase FDI

from other sources. Aid from both the EU (Model 4) and China (Model 8) leads to a positive

difference-in-difference that is significant at the 1% level. In both instances the magnitude is

larger than Model 2 in Table 1, with EU aid leading to a 35.7 percent increase in the

8

likelihood of FDI and Chinese aid increasing the likelihood of a non-Chinese FDI project by

41.4 percent. In contrast, aid from the US and Japan leads to no statistically significant

increase in the likelihood of an active site receive outside FDI. These results are consistent

with the literature above which suggests that the US and Japan may be more self-interested

donor agents, while the EU is more normative in its actions. The China result is also

consistent with building evidence that Chinese aid is broadly conducive to (immediate)

economic growth (Dreher et al. 2017).

Robustness

This section subjects the results above to a number of robustness checks. The full tables of

results can be found in the supplementary online appendix. The first check is to cross-

validate the result on a larger sample. AidData has geo-coded, project-level information for

Chinese and World Bank projects in all African countries. Accordingly, the analysis is

expanded to all of these locations, both with aggregated aid projects and via examining the

World Bank and China separately. The substantive results are maintained, with combined

Chinese and World Bank aid (Model S1) leading to a positive difference-in-difference in

active aid sites attracting FDI compared to inactive sites. However, the magnitude of results

above are only maintained for World Bank aid (Model S2) which increases active sites

chances of attracting FDI by 41.7 percent, a difference-in-difference significant at the 1%

level. While the difference-in difference of Chinese aid projects (Model 3) is positive and

significant at the 5% level, the magnitude is considerably smaller than the result from the

AIMS countries subsample (Model 8).

Second, the paper includes different sets of site-levels controls as well as lagging the aid

projects. Aid effectiveness literature has suggested that aid projects, particularly

infrastructure, may take some time to be realized (Brazys 2010; Bearce et al. 2013).

Accordingly, it may take some passage of time before a site becomes more attractive to FDI

and as such the robustness checks explore several lags (Models S4-S6). Likewise, to check

the robustness of the results, a model with no controls (S7), as well as a model with

additional controls (S8), including other measures of local corruption, including local

corruption perceptions of police, judges, tax officials, and MPs as well as local corruption

experiences of paying bribes for permits or to the police, measures of average household

cash flows, and reported experienced with discussing politics, are examined. The models

using lags and alternative controls are all substantively similar to the main results.

9

Table 2: Source Country Heterogeneity in African AIMS Countries

(1) (2) (3) (4) (5) (6) (7) (8) VARIABLES US FDI NonUS FDI EU FDI NonEU FDI JP FDI NonJP FDI CN FDI NonCN FDI

active_source 0.753*** 0.558*** 0.869*** 0.857*** (0.075) (0.033) (0.051) (0.034) inactive_source 0.682*** 0.587*** 0.781*** 0.793*** (0.032) (0.041) (0.036) (0.031) active_nonsource -0.334*** 0.019 -0.181*** -0.033 (0.054) (0.052) (0.045) (0.045) inactive_nonsource -0.348*** -0.338*** -0.252*** -0.447*** (0.051) (0.045) (0.058) (0.058) Observations 1,371 1,371 1,246 1,246 1,371 1,371 1,371 1,371 R-squared 0.263 0.219 0.320 0.222 0.271 0.143 0.312 0.199 Baseline controls YES YES YES YES YES YES YES YES Year FE YES YES YES YES YES YES YES YES Country FE YES YES YES YES YES YES YES YES Difference in difference 0.071 0.014 -0.029 0.357 0.088 0.071 0.064 0.414 F test: active-inactive=0 0.793 0.035 0.338 26.085 2.305 0.987 1.774 32.815 p value 0.373 0.851 0.561 0.000 0.129 0.321 0.183 0.000

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

10

Third, as discussed above, the primary analysis uses AfroBarometer enumeration sites from

which to assess proximity to projects. While using these sites has the advantage of using

aggregated, site-level, controls from that survey, the paper checks robustness of these

results by using generic grid cells, which have the distinct advantage of uniformly and

completely covering the spatial analysis area. The base grid-cell data comes from the

International Food Policy Research Institute (IFPRI 2017) and utilizes grids spaced at five-

minute distances. This data also contains variables on land usage including area of

cropland, urban area, and area of water bodies. Controlling for each of these variables, the

results are substantively reproduced looking at both 25km (Model S9) and 50km (Model

S10) distances. Model S11 also limits the analysis to only grid cells that have evidence of

habitation (crop or urban land usage), again with consistent results.

Finally, it is possible that the outcome process is spatially auto-regressive. The study of

economic geography is premised on the fact that economic concerns may cluster for a

variety of reasons, leading to spatial autocorrelation. This possibility is tested for by

calculating Moran’s I on the residuals from model II. As shown in the supplemental online

appendix (Figure SA1), the spatial correlation coefficient at close proximity (bands 1-2 and 1-

3) is not significantly different from the expected statistic under conditions of no spatial

correlation. However, when looking further afield (bands 1-4 and 1-5) while the correlation

coefficient is still quite small, it now is statistically different form the expected value,

indicating the presence of spatial autocorrelation. Arguably these results are more

suggestive of political boundary effects, rather than spatial autocorrelation, but as a further

robustness check the specification is run using both a spatial autoregressive (SAR) (Model

S12) and a spatial error model (SEM) (Model S13). In both models the difference-in-

difference between active and inactive sites is significant at the 5% level and the magnitude

is quite similar to that in Model 2 above. Moreover, the p values of rho (Model S12) and

lambda (model S13) suggest that any spatial autocorrelation is sufficiently addressed by the

models

Conclusions

This paper’s findings convincingly suggest that aid facilitates investment. By extension, the

result supports the claim that aid can facilitate local economic activity. Indeed, this finding is

compatible with several recent studies which identify a local impact of aid on growth (Civelli

et al. 2018; Bitzer and Goren 2018). Interestingly, the causal linkage between aid and

investment doesn’t hold when considering aid and FDI from the same source actor. Rather

than aid causing FDI in this instance, aid and FDI from the same donor appear to go hand-

11

in-hand, suggesting that both are driven by the same locational considerations, perhaps as a

packaged approach to foreign economic policy. However, aid from two major donor actors,

the EU and China, does also increase local FDI from other countries. In contrast, aid from

the US and Japan does not attract FDI from other sources. Regardless of if the EU and

Chinese results are an intended or unintended spillover from a more self-serving foreign

economic policy, it is suggestive that aid can build the public goods that attract FDI.

While these findings complement recent subnational analyses that finds that “aid works”, it

should be noted that nothing can be determined about local socio-economic impacts or

distributive effects. Few studies have yet linked FDI to local impacts and further work in this

direction is vital to understand if the aid-FDI linkage ultimately engenders positive local

development outcomes.

12

References

Addison, T., & Baliamoune-Lutz, M. (2016). The effects of Aid on Foreign Direct Investment in Africa and other regions. AidData. 2016a. DRC-AIMS_GeocodedResearchRelease_Level1_v1.3.1 geocoded dataset. Williamsburg, VA and Washington, DC: AidData. Accessed on [17-09-2018]. http://aiddata.org/research-datasets. AidData. 2016b. NigeriaAIMS_GeocodedResearchRelease_Level1_v1.3.1 geocoded dataset. Williamsburg, VA and Washington, DC: AidData. Accessed on [17-09-2018]. http://aiddata.org/research-datasets. AidData. 2016c. UgandaAIMS_GeocodedResearchRelease_Level1_v1.4.1 geocoded dataset. Williamsburg, VA and Washington, DC: AidData. Accessed on [17-09-2018]. http://aiddata.org/research-datasets. AidData. 2016d. SenegalAIMS_GeocodedResearchRelease_Level1_v1.5.1 geocoded dataset. Williamsburg, VA and Washington, DC: AidData. Accessed on [17-09-2018]. http://aiddata.org/research-datasets. AidData. 2017a. BurundiAIMS_GeocodedResearchRelease_Level1_v1.0 geocoded dataset. Williamsburg, VA and Washington, DC: AidData. Accessed on [17-09-2018]. http://aiddata.org/research-datasets AidData. 2017b. SierraLeoneAIMS_GeocodedResearchRelease_Level1_v1.0 geocoded dataset. Williamsburg, VA and Washington, DC: AidData. Accessed on [17-09-2018]. http://aiddata.org/research-datasets. Alesina, A., & Dollar, D. (2000). Who gives foreign aid to whom and why?. Journal of economic growth, 5(1), 33-63. Amusa, K., Viegi, N., & Monkam, N. (2016). The nexus between foreign direct investment and foreign aid: an analysis of sub-Saharan African countries. African Finance Journal, 18(2), 45-68. Anyanwu, J. C. (2012). Why Does Foreign Direct Investment Go Where It Goes?: New Evidence From African Countries. Annals of Economics & Finance, 13(2). Arazmuradov, A. (2015). Can Development Aid Help Promote Foreign Direct Investment? Evidence from Central Asia. Economic Affairs, 35(1), 123-136. Bearce, D. H., Finkel, S. E., Pérez-Liñán, A. S., Rodríguez-Zepeda, J., & Surzhko-Harned, L. (2013). Has the new aid for trade agenda been export effective? Evidence on the impact of US AfT allocations 1999–2008. International Studies Quarterly, 57(1), 163-170. BenYishay, A., Rotberg, R., Wells, J., Lv, Z., Goodman, S., Kovacevic, L., Runfola, D. 2017. Geocoding Afrobarometer Rounds 1-6: Methodology & Data Quality. AidData. Available online at http://geo.aiddata.org . Bermeo, S. B. (2017). Aid Allocation and Targeted Development in an Increasingly Connected World. International Organization, 71(4), 735-766. Berthélemy, J. C., & Tichit, A. (2004). Bilateral donors' aid allocation decisions—a three-dimensional panel analysis. International Review of Economics & Finance, 13(3), 253-274

13

Bitzer, Jurgen and Erkan Goren. (2018) Foreign Aid and Subnational Development: A Grid Cell Analysis. AidData Working Paper #55. Williamsburg, VA: AidData at William & Mary. Blair, Robert A., and Philip Roessler. The Effects of Chinese Aid on State Legitimacy in Africa: Cross-National and Sub-National Evidence from Surveys, Survey Experiments, and Behavioral Games. AidData Working Paper #59. Williamsburg, VA: AidData at William & Mary. Blaise, S. (2005). On the link between Japanese ODA and FDI in China: a microeconomic evaluation using conditional logit analysis. Applied Economics, 37(1), 51-55. Brazys, S. (2010). Race to give? The selective effectiveness of United States trade capacity building assistance. Review of International Political Economy, 17(3), 537-561. — (2013). Evidencing donor heterogeneity in Aid for Trade. Review of International Political Economy, 20(4), 947-978. Brazys, S., & Lightfoot, S. (2016). Europeanisation in Aid for Trade: the impact of capacity and socialisation. European Politics and Society, 17(1), 120-135. Brazys, S., Elkink, J. A., & Kelly, G. (2017). Bad neighbors? How co-located Chinese and World Bank development projects impact local corruption in Tanzania. The Review of International Organizations, 1-27. Brazys, S., & Dukalskis, A. (2017). Canary in the coal mine? China, the UNGA, and the changing world order. Review of International Studies, 43(4), 742-764. Brazys, S., & Regan, A. (2017). The politics of capitalist diversity in Europe: Explaining Ireland’s divergent recovery from the Euro crisis. Perspectives on Politics, 15(2), 411-427. Briggs, R. C. (2017). Does foreign aid target the poorest?. International Organization, 71(1), 187-206. Carbone, M. (2013). Between EU actorness and aid effectiveness: the logics of EU aid to sub-Saharan Africa. International Relations, 27(3), 341-355. Civelli, A., Horowitz, A., & Teixeira, A. (2018). Foreign aid and growth: A Sp P-VAR analysis using satellite sub-national data for Uganda. Journal of Development Economics, 134, 50-67 Cleeve, E. A., Debrah, Y., & Yiheyis, Z. (2015). Human capital and FDI inflow: An assessment of the African Case. World Development, 74, 1-14. Donaubauer, J. (2014). Does foreign aid really attract foreign investors? New evidence from panel cointegration. Applied Economics Letters, 21(15), 1094-1098. Donaubauer, J., Herzer, D., & Nunnenkamp, P. (2014). Does aid for education attract foreign investors? An empirical analysis for Latin America. The European Journal of Development Research, 26(5), 597-613. Donaubauer, J., Meyer, B., & Nunnenkamp, P. (2016). Aid, infrastructure, and FDI: Assessing the transmission channel with a new index of infrastructure. World Development, 78, 230-245.

14

DREHER, Axel; LOHMANN, Steffen. Aid and growth at the regional level. Oxford Review of Economic Policy, 2015, vol. 31, no 3-4, p. 420-446. Dreher, Axel, Andreas Fuchs, Bradley Parks, Austin M. Strange, and Michael J. Tierney. 2017. Aid, China, and Growth: Evidence from a New Global Development Finance Dataset. AidData Working Paper #46. Williamsburg, VA: AidData at William & Mary. Garriga, A. C., & Phillips, B. J. (2014). Foreign aid as a signal to investors: Predicting FDI in post-conflict countries. Journal of Conflict Resolution, 58(2), 280-306. Gil-Pareja, S., Vivero, R. L., & Paniagua, J. (2013). The effect of the great recession on foreign direct investment: global empirical evidence with a gravity approach. Applied Economics Letters, 20(13), 1244-1248. Hall, S. (2011). Managing tied aid competition: Domestic politics, credible threats, and the Helsinki disciplines. Review of International Political Economy, 18(5), 646-672. Harrigan, J., & Wang, C. (2011). A new approach to the allocation of aid among developing countries: is the USA different from the rest?. World Development, 39(8), 1281-1293. Hernandez, D. (2017). Are “New” Donors Challenging World Bank Conditionality?. World Development, 96, 529-549. Hook, S. W., & Zhang, G. (1998). Japan's aid policy since the cold war: rhetoric and reality. Asian Survey, 38(11), 1051-1066. Isaksson, A. S., & Kotsadam, A. (2018). Chinese aid and local corruption. Journal of Public Economics, 159, 146-159. Kimura, H., & Todo, Y. (2010). Is foreign aid a vanguard of foreign direct investment? A gravity-equation approach. World Development, 38(4), 482-497. Knutsen, C. H., Kotsadam, A., Olsen, E. H., & Wig, T. (2017). Mining and local corruption in Africa. American Journal of Political Science, 61(2), 320-334. Lee, H. H., & Ries, J. (2016). Aid for Trade and Greenfield Investment. World Development, 84, 206-218. McKinley, R. D., & Little, R. (1977). A foreign policy model of US bilateral aid allocation. World Politics, 30(1), 58-86. — (1979). The US aid relationship: a test of the recipient need and the donor interest models. Political Studies, 27(2), 236-250. Martorano, B., Metzger, L., & Sanfilippo, M. (2018). Chinese development assistance and household welfare in sub-Saharan Africa. UNU-MERIT Working Paper Series, 2018. Owen, E. (2018). Foreign Direct Investment and Elections: The Impact of Greenfield FDI on Incumbent Party Reelection in Brazil. Comparative Political Studies, 0010414018797936. Peratsakis, Christian, Joshua Powell, Michael Findley, Justin Baker and Catherine Weaver. 2012. Geocoded Activity-Level Data from the Government of Malawi's Aid Management Platform. Washington D.C. AidData and the Robert S. Strauss Center for International Security and Law.

15

Selaya, P., & Sunesen, E. R. (2012). Does foreign aid increase foreign direct investment?. World Development, 40(11), 2155-2176. Strange, Austin M., Axel Dreher, Andreas Fuchs, Bradley Parks, and Michael J. Tierney. 2017. Tracking Underreported Financial Flows: China’s Development Finance and the Aid–Conflict Nexus Revisited. Journal of Conflict Resolution 61 (5): 935-963. Su, H., Chen, Q., & Han, B. (2017). An Empirical Analysis on the Relationship between Chinese Outward Foreign Aid and Outward Foreign Direct Investment from China. Journal of China Studies, 20(3), 47-68.

16

Supplementary Online Appendix

Table SA1: Data Sources and Summary Statistics

Variable Variable Source Max Min Mean Std Dev. Observatio

ns

Cluster Level at 50km (Model 2)

FDI www.fdimarkets.com 1 0 0.590 0.492 1,371 Active www.aiddata.org 1 0 0.023 0.151 1,371 Inactive www.aiddata.org 1 0 0.028 0.166 1,371 Corruption Government

Officials BenYishay et al. 2017 http://geo.aiddata.org

http://www.afrobarometer.org

3 0 1.561 0.461 1,371

Police BenYishay et al. 2017 http://geo.aiddata.org

http://www.afrobarometer.org

3 0 1.837 0.540 1,371

Judges BenYishay et al. 2017 http://geo.aiddata.org

http://www.afrobarometer.org

3 0 1.431 0.427 1,371

Tax Officials BenYishay et al. 2017 http://geo.aiddata.org

http://www.afrobarometer.org

3 0 1.604 0.472 1,215

MP BenYishay et al. 2017 http://geo.aiddata.org

http://www.afrobarometer.org

3 0 1.473 0.471 1,215

Permit Bribe BenYishay et al. 2017 http://geo.aiddata.org

http://www.afrobarometer.org

3 0 0.298 0.330 1,371

Police Bribe BenYishay et al. 2017 http://geo.aiddata.org

http://www.afrobarometer.org

3 0 0.317 0.384 1,371

17

Without Cash BenYishay et al. 2017 http://geo.aiddata.org

http://www.afrobarometer.org

1 0 0.676 0.268 1,215

Active www.aiddata.org 1 0 0.370 0.483 1.371 Inactive www.aiddata.org 1 0 0.139 0.346 1.371

Grid Cell Level at 50km (Model 10)

FDI www.fdimarkets.com 1 0 0.131 0.338 18,143 Active www.aiddata.org 1 0 0.123 0.329 18,143 Inactive www.aiddata.org 1 0 0.070 0.256 18,143 Urban www.ifpri.org 1 0 0.061 0.240 18,143 Crop www.ifpri.org 1 0 0.781 0.413 18,143 Area Water www.ifpri.org 8449.58 0 81.47 659.47 17,875

18

Table SA1: World Bank and China Pan-African Results, Lags, and Alternative Controls

(S1) (S2) (S3) (S4) (S5) (S6) (S7) (S8) VARIABLES WB and

China World Bank China Aid Lag 1 Aid Lag 2 Aid Lag

3 No Controls Expanded

Controls

active -0.179*** 0.059*** -0.174*** -0.124* -0.096 -0.124* -0.146** -0.109 (0.014) (0.014) (0.014) (0.075) (0.085) (0.075) (0.075) (0.079) inactive -0.216*** -0.358*** -0.220*** -0.407*** -0.406*** -0.407*** -0.426*** -0.364*** (0.018) (0.018) (0.018) (0.073) (0.073) (0.073) (0.072) (0.081) Observations 8,756 8,756 8,756 1,371 1,371 1,371 1,374 1,215 R-squared 0.331 0.345 0.331 0.234 0.233 0.234 0.216 0.240 Baseline controls YES YES YES YES YES YES NO NO Year FE YES YES YES YES YES YES YES YES Country FE YES YES YES YES YES YES YES YES Difference in difference

0.037 0.417 0.046 0.283 0.310 0.283 0.279 0.255

F test: active-inactive=0

2.845 351.085 4.300 7.612 7.842 7.612 7.560 5.275

p value 0.092 0.000 0.038 0.006 0.005 0.006 0.006 0.022

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

19

Table SA2: Grid Cell Approach

(S9) (S10) (S11) VARIABLES 25 km 50 km Inhabited Cells Only 50km

active -0.016*** -0.055*** -0.065*** (0.005) (0.006) (0.007) inactive -0.088*** -0.214*** -0.234*** (0.007) (0.011) (0.012) Observations 17,902 17,875 14,208 R-squared 0.066 0.211 0.210 Baseline controls YES YES YES Year FE YES YES YES Country FE YES YES YES Difference in difference 0.071 0.159 0.168 F test: active-inactive=0 89.150 164.160 146.227 p value 0.000 0.000 0.000

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

20

Figure SA1: Moran's I spatial correlogram

Residuals --------------------------------------------------------------

Distance bands | I E(I) sd(I) z p-value* --------------------+-----------------------------------------

(1-2] | 0.000 -0.001 0.005 0.238 0.406 (1-3] | 0.002 -0.001 0.003 0.753 0.226 (1-4] | -0.023 -0.001 0.003 -8.161 0.000 (1-5] | -0.024 -0.001 0.002 -9.855 0.000

-------------------------------------------------------------- *1-tail test

21

Table SA3: Spatial Autoregressive and Spatial Error Models

(S12) (S14) VARIABLES Spatial Lag Spatial Error

active -0.178** -0.184** (0.083) (0.086) inactive -0.430*** -0.427*** (0.074) (0.075) Observations 1,371 1,371 Baseline controls YES YES Year FE YES YES Country FE YES YES p value rho/lambda 0.908 0.768 Difference in difference 0.252 0.243 Chi2 test: active-inactive=0 5.303 4.534 p value Chi2 0.021 0.033

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1